From 05d2ebee5e4ebc31cd71f6064ca433a58692be76 Mon Sep 17 00:00:00 2001 From: Mithun Iyer Date: Mon, 14 Sep 2020 14:31:02 -0700 Subject: Test RST handling in TIME_WAIT. gVisor stack ignores RSTs when in TIME_WAIT which is not the default Linux behavior. Add a packetimpact test to test the same. Also update code comments to reflect the rationale for the current gVisor behavior. PiperOrigin-RevId: 331629879 --- pkg/tcpip/transport/tcp/rcv.go | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'pkg/tcpip') diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index bc920a03b..cfd43b5e3 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -436,6 +436,13 @@ func (r *receiver) handleTimeWaitSegment(s *segment) (resetTimeWait bool, newSyn // Just silently drop any RST packets in TIME_WAIT. We do not support // TIME_WAIT assasination as a result we confirm w/ fix 1 as described // in https://tools.ietf.org/html/rfc1337#section-3. + // + // This behavior overrides RFC793 page 70 where we transition to CLOSED + // on receiving RST, which is also default Linux behavior. + // On Linux the RST can be ignored by setting sysctl net.ipv4.tcp_rfc1337. + // + // As we do not yet support PAWS, we are being conservative in ignoring + // RSTs by default. if s.flagIsSet(header.TCPFlagRst) { return false, false } -- cgit v1.2.3